In other words, the word embeddings can be dreadfully sexist. This happens because any bias in the articles that make up the Word2vec corpus is inevitably captured in the geometry of the vector space.